Ten Projects, One Person: What I Keep, What I Pause, What I Kill

Product · 2026-08-11

I've got ten public projects running at the moment. Eight are live, one's in development, one's paused on purpose. That last bit is deliberate.

Anyway, I want to talk about how I decide which of these gets my actual attention, which ones get left to tick along quietly, and which ones get killed outright before they cost me anything real. Because the honest answer isn't "I'm just really productive". It's more boring than that, and also more useful.

The two that get my attention

Most of the ten don't need me most days. Two do: ReviewNudge and ScriptGrain.

ReviewNudge sends SMS review requests for local businesses, simple job, does it well. ScriptGrain writes AI content that actually sounds like you wrote it, not a robot doing an impression of you. Both have real users. That's the whole reason they get my time and the other eight, mostly, don't.

I used to think focus meant picking one thing and killing everything else. Turns out that's not quite right. You can run ten things if the nine quiet ones aren't asking anything of you. The two loud ones will always ask something of you though, and that's fine, that's the job. Users mean feedback, feedback means changes, changes mean me, actually me, sat there doing the work.

How ten things don't collapse into chaos

Here's the bit that sounds least believable. I've got a fleet of over 20 Claude agents running under something I call Jarvis, still in development itself funnily enough, and they do the maintenance work I'd otherwise be drowning in. Audits, monitoring, content drafting with a QA pass before anything goes near a user. Not glamorous, but it's what stops a product quietly rotting.

That's what lets a small data product like the GLA Spend Explorer, a tool for poking around London public spending, just sit there quietly working without rotting. Nobody's manually checking it every week. It idles cheaply, because the fleet checks on it so I don't have to.

The constraint isn't tools, or time really, it's attention. And attention goes where the users are shouting loudest. Which, right now, is ReviewNudge and ScriptGrain.

The pause: Football Scout Engine

This one I actually loved building.

Football Scout Engine used computer vision for football scouting, proper research, genuinely went well. The models worked. The approach was sound. Then I hit a wall that had nothing to do with code.

The sports research datasets I'd been using are licensed for benchmarking only. You can look, you can test, you can publish papers about it. You cannot ship a commercial product trained on that data. I checked. Checked again. There wasn't a clever workaround here, just a licence that meant no.

I could've fudged it. Found a grey area, hoped nobody asked too many questions, shipped something anyway. I didn't. I paused it instead, and I show it publicly as paused, not quietly archived somewhere nobody looks.

My studio's called Adapt Progress Evolve. Adapt includes knowing when to stop. That's not a slogan, it's just true, and I'd rather live by the boring version of it than the exciting version where I ignore a licensing problem and hope it goes away.

The kill: the stock-checker that never got built

This is the one I'm actually proud of, weirdly.

Had an idea for a stock-checker app, something that would tell you in real time what's actually on the shelf at your local supermarket before you order a delivery. Good idea, I thought. So before writing a line of code, I had the agents properly research it.

Verdict came back red. The supermarket delivery platforms the whole idea depended on keep their APIs closed to third parties. There was no legal way in. No workaround, nothing. The product's entire premise needed data I could never actually get.

So I killed it. Dead, gone, no product code, nothing to show for it except a research report.

And honestly, that felt great. Not in a smug way, in a relieved way. Because the alternative universe where I don't research first is the one where I spend three months building something on data that was never mine to use, then find out the hard way. A killed idea that cost a research report is nothing. A killed idea that cost three months is a proper wound.

Why I show any of this at all

I don't trust a track record with no scar tissue in it.

Anyone can show you eight live products and call it a portfolio. It's easy to make. What's harder, and more honest, is showing the paused project sat there next to them, and the killed idea that never made it past a document. That's the bit that tells you something true about how decisions actually get made here.

So that's the lot. Ten projects, two getting my attention because that's where the users are, one paused because the licence said stop and I listened, plus one idea that never existed past a report, which is exactly as it should be.